文本情感分析（一）

最近和小伙伴一起研究文本的情感分类，所以做一个记录。

一、背景：

       商品的评论信息，好评与差评约各20000条。

二、目标：

        根据现有的样本做一个分类器。

三、方案：

        方案一：情感类词库，进行线性求和。

        方案二：句向量+传统机器学习分类器。

        方案三：句向量+DNN分类器

        方案四：词向量+lstm分类器

直接上代码：此次更新的代码为方案二与三，其他后续更新

# import modules & set up logging
import gensim,logging
import pandas as pd #导入Pandas
import numpy as np #导入Numpy
import jieba #导入结巴分词
from __future__ import absolute_import #导入3.x的特征函数
from __future__ import print_function


neg=pd.read_excel('neg.xls',header=None,index=None)
pos=pd.read_excel('pos.xls',header=None,index=None) #读取训练语料完毕
pos['mark']=1
neg['mark']=0 #给训练语料贴上标签
pn=pd.concat([pos,neg],ignore_index=True) #合并语料
cw = lambda x: list(jieba.cut(x)) #定义分词函数
pn['words'] = pn[0].apply(cw)

sentences = pn['words']
model = gensim.models.Word2Vec(sentences,size=64,window=5, min_count=0,workers=4)

idx = range(len(pn))
np.random.shuffle(idx)
pn = pn.loc[idx]

def sen2vec(pd_word):
    gg=[]
    for i in pd_word.values:
        a=np.zeros(64)
        n=0
        for j in i:
            for k in j:
                a += model[k]
                n=n+1
        gg.append(a/n)
    return gg

pn['s2v']=sen2vec(pn.iloc[:,[2]])

from sklearn.svm import SVC,LinearSVC,NuSVC
from sklearn.naive_bayes import MultinomialNB,BernoulliNB
from sklearn.linear_model import LogisticRegression
from sklearn.ensemble import RandomForestClassifier
from sklearn.metrics import accuracy_score
from sklearn.metrics import roc_auc_score
from sklearn.svm import SVC,LinearSVC,NuSVC
from sklearn.ensemble import GradientBoostingClassifier

train_num = 15000
#按keras的输入要求来生成数据
x = np.array(list(pn['s2v']))
y = np.array(list(pn['mark']))
y = y.reshape((-1,1)) #调整标签形状

def model_score(modelname,x,y,train_num):
    model = modelname
    model.fit(x[:train_num],y[:train_num])
    y_predict = model.predict(x[train_num:])
    (acc,roc) = accuracy_score(y[train_num:],y_predict),roc_auc_score(y[train_num:],y_predict)
    return (acc,roc)

print('BernoulliNB`s accuracy is %f, auc_score is %f' % model_score(BernoulliNB(),x,y,train_num))
print('LogisticRegression`s accuracy is %f, auc_score is %f' % model_score(LogisticRegression(),x,y,train_num))
print('RandomForestClassifier`s accuracy is %f, auc_score is %f' % model_score(RandomForestClassifier(),x,y,train_num))
print('LinearSVC`s accuracy is %f, auc_score is %f' % model_score(LinearSVC(),x,y,train_num))
print('GradientBoostingClassifier`s accuracy is %f, auc_score is %f' % model_score(GradientBoostingClassifier(),x,y,train_num))

from keras.models import Sequential
from keras.layers import Dense, Activation, Dropout, Embedding,Flatten
from keras.layers import LSTM,GRU,Conv2D,MaxPooling2D,Conv1D,GlobalAveragePooling1D,MaxPooling1D
from keras.optimizers import SGD

#MLP for binary classification
model = Sequential()
model.add(Dense(64,input_dim=64,activation='relu'))
model.add(Dropout(0.5))
model.add(Dense(64,activation='relu'))
model.add(Dropout(0.5))
model.add(Dense(1,activation='sigmoid'))

model.compile(optimizer='rmsprop',
              loss='binary_crossentropy',
              metrics=['accuracy'])
model.fit(x[:train_num],y[:train_num],epochs=100,batch_size=32)

score =model.evaluate(x[train_num:],y[train_num:],batch_size=16)
score

结果如下：

BernoulliNB`s accuracy is 0.658640, auc_score is 0.658055
LogisticRegression`s accuracy is 0.767731, auc_score is 0.767476
RandomForestClassifier`s accuracy is 0.808190, auc_score is 0.808713
LinearSVC`s accuracy is 0.774120, auc_score is 0.773867
GradientBoostingClassifier`s accuracy is 0.786405, auc_score is 0.786367
dnn`s accuracy is 0.824570024

结论，相同情况下：

1、可以看到传统的机器学习随机森林较为明显。

2、dnn网络的话表现还是比较好的。